Sign in with Google

Open-source AI coding agents to try first

Coding agents you can read, change and run with the model you choose, in the terminal, the editor or the browser.

For developers who want an AI agent they can read, change and point at any model. We picked one or two strong agents for each place you work: the terminal, the editor, the browser and the desktop.

8 picks in 4 sections · checked by an editor on Oct 30, 2026 · How we pick

01Terminal5 picks

  1. OpenCodeanomalyco/opencode

    The most popular open terminal agent. Works with almost any model provider.

    Desktop · CLI · MIT · active 23 hours ago · ★ 224k

    96worth
  2. Codex CLIopenai/codex

    Pick this if you already pay for ChatGPT and want its agent in your terminal.

    CLI · Apache-2.0 · active 2 days ago · ★ 133k

    92worth
  3. Gemini CLIgoogle-gemini/gemini-cli

    Pick this for a generous free tier with a Google account.

    CLI · Apache-2.0 · active 2 days ago · ★ 109k

    91worth
  4. Qwen CodeQwenLM/qwen-code

    Pick this to run a strong agent with open-weight Qwen models.

    CLI · Apache-2.0 · active 23 hours ago · ★ 29k

    88worth
  5. Piearendil-works/pi

    A small agent you can read in an afternoon and extend yourself.

    CLI · Library / SDK · MIT · active yesterday · ★ 119k

    92worth

02Editor1 pick

  1. Clinecline/cline

    Pick this if you want the agent inside VS Code, asking before each step.

    Extension / plugin · Apache-2.0 · active 2 days ago · ★ 71k

    88worth

03Desktop1 pick

  1. gooseaaif-goose/goose

    Pick this for a desktop app that also does work outside code.

    Desktop · CLI · Apache-2.0 · active 2 days ago · ★ 56k

    91worth

04Browser1 pick

  1. OpenHandsOpenHands/OpenHands

    Pick this to run agents in a sandbox and give them whole tickets.

    Web · Docker / self-host · CLI · MIT · active yesterday · ★ 92k

    89worth

More in AI tools by job

Add-ons that make your coding agent better

Memory, token savers, specs and skills that make Claude Code, Codex and other coding agents faster and cheaper.Inside: Claude-Mem, RTK, Headroom + 7 more10 picks · 6 sections

AI tools for testing web apps and AI features

Let an AI agent click through your web app, write browser tests, or check the answers your own AI feature gives.Inside: Playwright MCP, playwright-cli, Chrome DevTools MCP + 7 more10 picks · 2 sections

AI tools for UI and product design

Design apps, turn screenshots into code and give your coding agent good taste, with open-source tools.Inside: Open Design, open-pencil, Onlook + 5 more8 picks · 3 sections

Other groups

See every project in this category on Explore →Missing a project? Suggest it →

How we pick

An editor picks 6 to 12 projects for one job and writes when to choose each one. Every pick had a commit in the last 6 months. It is not archived. Its worth score is 60 or more. An editor breaks a rule only for a reason, and writes that reason in the pick's note. We check every list again each quarter. A pick goes when it breaks a rule or a clearly better project appears.